"This powerpoint presentation is property of David Abbink and Delft University of Technology. No part of this publication may be reproduced, stored in other retrieval systems or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior permission (mail d.a.abbink@tudelft.nl for any requests).
Algemene Leden Vergadering PLUS Delft, 24 mei 2016
Programme Welcome 10.00-10.45 David Abbink PhD, Associate Professor at the Department of BioMechanical Engineering, Delft University of Technology. Why should autonomous cars be like horses? 10.45-11.30 Marly de Blaeij, Policy Advisor at Verbond Van Verzekeraars. Keep on rollin 11.30-12.00 AGM 12.00-14.00 Walking lunch
Simulation The future of intelligent vehicles Self-driving or co-operative driving?
Who am I? Dr. ir. David Abbink - Associate Professor of BioMechanical Engineering Delft University of Technology, The Netherlands Delft Haptics Lab: visiting prof.: 0.1 fte (Erwin Boer) postdoc: 1 postdoc (+1) PhDs: 6 PhDs (+1) www.delfthapticslab.nl MSc students: 8 / year TU Delft Funding Nissan (2002 2015), Boeing (2007 2011) Dutch Science Foundation NWO-STW VENI (2011), VIDI (2015), H-Haptics (2011-2017)
Humans and the tools we make
Ray Kurzweil & Moore s law
Machine Learning & Deep Learning
Artificial intelligence beats us already!
Simulation Learning AI - Deep learning
Embodiment: What is intelligence without movement?
Physical motion is difficult! Darpa Robotics Challenge, 2015
Or should we start to worry?
Simulation Robots will kill us all!
The human robot evolution
What are the risks and who s responsible?
Co-existence and co-operation
Enough stupid interaction with intelligence
How do we control our body? Sixth sense
The future of driving? 1955 2015
Yeah right, what about self-driving cars?
Yeah right, what about self-driving cars?
What s wrong with highly automated driving? 2014: impressive technology.. outdated interaction! Automated car has limitations driver remains an essential component! Inhuman Task! Traded Control, human-as-backup
What s wrong with highly automated driving? OK my hands are near the wheel, my foot s near the brake, but I m not touching them this is hands-free driving but not the vision of mind-free driving Therefore: Slow (25 mph) Driver remains responsible Research focus on warning systems and driver monitoring Owwwsh*t! -This might not correspond to your driving style 1. How to make the automation understand your preferences/abilities? Oh that was a cone! 2. How to allow the driverto understand automation boundaries and limitations, and react on time?
What about trust?
What about conflicts?
My approach: creating a feel for the robot
My approach: creating a feel for the robot Haptic Gas Pedal Pedal Force Own car Pedal Depression
Haptic Shared Control for Steering Steering Wheel T Steering Torque Can generate feedback forces but: driver can relax, resist or give way X opt 0 Steering X sw Angle
Delft Approach to Design of Human-centered Haptic Shared Control Abbink & Mulder (2009) Exploring the dimensions of haptic feedback support in manual control Steering Wheel T Joint patent with Nissan (2008) Can generate feedback forces Can modify impedance dynamically shift authority in changing criticality X opt 0 X sw
Single path vs Multiple Paths One approach to support multiple paths? How to support lane changes? Tsoi et al. (2010) IEEE SMC Conference How to support multiple evasive paths? Della Penna et al. (2010) IEEE SMC Conference Ideally, human should make the choice Creative solutions may be needed Liability
Design Concept for multiple paths Reduce stiffness - criticality will be felt when trying to steer - easier to steer left or right T X opt 0 X opt X sw
Design Concept for multiple paths Stiffness T Can become negative in extreme cases - a choosing human is supported to avoid obstacle, and is then caught by the support - a stubborn human needs to increase own stiffness to avoid steering left or right X opt 0 X sw
Manual vs Shared Mulder, Abbink & Boer (2012) - Sharing Control with Haptics - Seamless Driver Support from Manual to Automatic Control Human Factors Tested 3 driver groups (from young and unexperienced, to old and experienced), during curve negotiation in a fixed-base driving simulator. Goal: compare manual control, to shared control, to hands-free driving Safety margins increase with less effort
Classic automation vs shared control Flemisch et al (2008) Method: Test automation errors of a curve negotiation support system that would fail just before the onset of a sharp curve Conditions with full automation (red lines) that allowed manual override with haptic shared control (green lines)
Nominal Mode Different Designs of Haptic Shared Control vs Critical Failure Petermeijer, Abbink, de Winter (2014) Human Factors Prize 2014
My goal: develop Symbiotic Driving VENI: One-size-fits-all & One-size-fits-always: VIDI: Symbiotic interaction : system adapts like a human, and adapts to driver Visual feedback Force feedback Physical Interaction Control input Sensor feedback
Implications Haptic Shared Control is a unified approach Continuous sharing of control through forces No binary switches (on/off), but smooth shifting of authority Drivers are continuously engaged (and need to be so!) Driver is better aware of the functionality and intent of the system, as well as changing criticality of situation Drivers can always overrule the system Physically links driver to respond to driving environment, allowing fast reflexes and neuromuscular adaptation It enables mutual learning and adaptation between the driver and the intelligent vehicle Keeps the driver in the loop, comfortably responsible and aware
What are the risks and who s responsible?
The future is co-operation TU Delft
d.a.abbink@tudelft.nl www.delfthapticslab.nl Vragen? Collaborators Mark Mulder Max Mulder Erwin Boer Joost de Winter MSc Students